Products·3 min read
By BitsMindsSource: SemiAnalysis

Cerebras CS-4 Doubles Speed on the Same 5nm Chip

The CS-4 puts three wafer-scale engines in one rack for 750 petaflops and more than 4,400 tokens a second per user — using the same silicon as the CS-3, clocked roughly twice as fast.

CEREBRAS 3 × WSE-3T · ONE RACK 750 PFLOPS 4,400 tokens/sec per user 2× the CS-3 BITSMINDS.COM
Share:

Cerebras unveiled the CS-4 on Tuesday, a rack-scale inference system the company says roughly doubles the throughput of its predecessor without moving to a new process node or a new chip. The three processors inside it are the same 5-nanometer wafer-scale engines that shipped in the CS-3 — this time running at close to twice the clock speed.

Each Wafer Scale Engine 3 Turbo carries four trillion transistors and 900,000 AI-optimized cores across 46,225 square millimetres of silicon: a complete TSMC wafer rather than a die cut from one. Packing three of them into a single rack gives the CS-4 750 petaflops of AI compute, 129.6 petabytes per second of aggregate on-wafer memory bandwidth, and 7.2 terabits per second of off-wafer I/O — double the per-wafer 2.4Tb/s link the CS-3 offered.

The gain comes from power and packaging rather than lithography. Cerebras rebuilt the chassis around a modular “backpack” that pulls power delivery out of the compute plane, letting the company push far more current into each wafer and pull the heat back out. A CS-4 rack draws 125 to 135 kilowatts, roughly twice a CS-3, while using about 50% fewer components. Cerebras also claims up to ten times more throughput per watt than the CS-3 — a figure that sits awkwardly next to a doubling of both speed and rack power, and one the company has not tied to a published workload.

On measured numbers the system is genuinely quick. Running GPT-OSS-120B, Cerebras reports more than 4,400 output tokens per second for a single user, against roughly 2,000 on the CS-3, and claims speeds up to thirty times faster than GPU-based setups in specific configurations. Direct Wafer Links between the three engines cut inter-processor latency from about five microseconds to two, which is what lets a three-wafer box behave like one accelerator instead of a small cluster.

The constraint Cerebras did not fix is memory. On-wafer SRAM stays at 44GB per engine, unchanged from the CS-3, and that ceiling is what limits long-context inference: production deployments still have to pair the CS-4 with HBM-based hardware in a disaggregated setup to hold large key-value caches. Speed per user is the CS-4’s pitch; capacity per user is still somebody else’s job.

First deliveries are scheduled for this quarter, and Cerebras has not published pricing. The company says a follow-on generation could arrive as early as 2027, targeting four times the speed and twenty times the throughput of today’s system by the end of that year. That roadmap lands in a market that has shifted decisively from training to serving — and for a company that has had to defend its valuation since a rocky IPO debut in May, an inference machine that ships this quarter is a more useful argument than a chip that ships in two years.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles

Businesses in ChatGPT Editorial illustration on an emerald field: a small storefront tilted toward the viewer, with a mint and white striped awning, a door and a window, a shop window carrying the OpenAI logo in white and a blue scalloped verified badge. The type column reads OpenAI beta, Businesses in ChatGPT, profiles, agents, insights. An editorial rendering, not a screenshot of the product. OPENAI · BETA Businesses in ChatGPT PROFILES · AGENTS · INSIGHTS BITSMINDS.COM
Products

Businesses in ChatGPT: OpenAI Opens Profiles and Agents

Claude Sheets and Claude Whiteboards Editorial illustration on a sage green field: a framed whiteboard tilted toward the viewer, covered in pastel sticky notes, with a small spreadsheet card clipped to its corner and a round clay magnet carrying the Claude logo. The type column reads Claude, new artifacts, Sheets, Whiteboards, Data, Meetings next. An editorial rendering, not a screenshot of the product. CLAUDE · NEW ARTIFACTS Sheets Whiteboards DATA · MEETINGS NEXT BITSMINDS.COM Claude
Products

Claude Sheets and Whiteboards Launch; Meetings Is Next

Google TPU 8tx and TPU 8i Max Editorial illustration on a deep blue field: a lit, slightly tilted green accelerator board with a row of heatsink fins, two gold-die chip packages and a raised white heat spreader in the middle carrying the Google G logo. The type column reads Google Cloud eighth-generation TPU, TPU 8tx, 8i Max, confidential and long context. An editorial rendering, not a photograph of Google hardware. GOOGLE CLOUD · 8TH-GEN TPU TPU 8tx 8i Max CONFIDENTIAL · LONG CONTEXT BITSMINDS.COM
Products

Google TPU 8tx and 8i Max Join the Eighth-Gen Lineup